Intel Skylake / Kaby Lake

[Edrick]

That looks very good. What specifically did you change? So now we know Skylake-X has full performance of M.2 when CPU PCIe lanes are used. Could you also re-test M.2 on PCH?

On my old box (6600K & Z170x), any little task, including SSD benchmarks, would ramp up the CPU to full speed even using the "Balanced" windows power setting.

With my new box (7820X and X299), I need to get at least 3 tasks running before CPU fully ramps up. Even running Prime95 on just 2 threads will not do the trick. It seems that these CPUs are more geared towards server power saving mode. Simply changing to "Performance" mode in windows and setting Min CPU speed to 100% was enough to give proper results in the SSD benchmarks.

Honestly, I thought that the new Intel SpeedShift technology (on Skylake) was supposed to improve the CPU EIST penalty. Not so much.

 

[dullard]

Interfaces made of polymer paste will fail eventually.

Paste fails too, but at what, maybe a ~10X longer time period? I honestly don't know how much longer, but it is significantly longer.

Please elaborate on the die size or/ and design life expectancy where this becomes an issue under ambient cooling.

"The 20/80°C reliability test results suggested that the failure mechanism of SAC assemblies is similar to that of conventional ATC profiles (0/100°C, -40/125°C) commonly performed in industry." http://ieeexplore.ieee.org/document/6575693/?reload=true . Conclusion: the below ambient 0°C and -40°C cycling didn't have any difference on solder failure.

As for area: try starting reading here (one of many papers): https://engineering.lamar.edu/_files/documents/mechanical/dr.-fan-publications/2008/Fan 2008_13 ECTC_3.pdf
Look at equation 9, page 140. Ep = the potential cyclic fatigue damage. The potential for cyclic fatigue damage is inversely proportional to A, the effective solder bond area. The larger the area, the lower the potential for cyclic fatigue damage.

Okay, I did your bidding. Now please elaborate, with proof, on how soldering is not bad for modern quickly heating and small CPUs.

 

[.vodka]

Lifetime reliability.

http://overclocking.guide/the-truth-about-cpu-soldering/

I think those >13yr old soldered Pentium 4s and Ds are running just fine out there. Same for all the 10yr old C2D E6xxx, E8xxx, Q6xxx, Q9xxx, and more recently Nehalems and Sandies.

Cedar Mill Pentium 4/Ds are 81mm², Conroe is 143mm², Wolfdale is 111mm², Nehalem/Lynnfield are 263/296mm², Sandy is 216mm²... not to mention the EP and EX versions of the latter two, Nehalem-EX is a 684mm² behemoth. There you have a mix of small and big dies throughout the years that have been soldered in the past without much issue, in many different packages.

Have you heard of any of these outright failing? It was most probably caused by a heatsink completely full of dirt and neglect.

What you had in the past was TIM in the lower end processors like E2xxx, E4xxx, E5xxx, E7xxx, Q8xxx C2Ds... the higher end models were soldered. Hell, Intel knew why they soldered Clarkdale's CPU and left the MCH+GPU with a thermal pad. I mean why even bother with solder here? Performance reasons..? Sure, this is probably why we started seeing >4GHz overclocks on Clarkdale when it came out, and it would only get better with Sandy.

That 32nm node was awesome! Those processors wouldn't have been so popular if they were TIM'd. You'd probably not heard of 5GHz stable done on Sandy silicon lottery winners without delidding.


Reliability isn't a problem with soldering in the time spans we're seeing so far (15yrs at most going back to Pentium 4/D). Money, materials and bean counting are, starting with Ivy Bridge back in 2012 and these concerns lately being stretched to the point of being ridiculous with Skylake-X. Some points der8auer raises in his analysis are... doubtful. It's a great piece nonetheless. Is the material stack needed to solder seen in der8auer's piece in short supply? AMD seems not to think so, as they've been soldering non stop since A64 came out and Ryzen is soldered. Sure Intel handles much more volume, but still.

It was alright when the high end models were soldered and the cheap stuff wasn't. This is a move that probably makes sense for them in the financial side of things, but on the other hand and going to a more recent example... the heat produced by those AVX512 units would pretty much prefer solder to TIM in whatever environment those processors are deployed in.

 

[StefanR5R]

@dullard, equation 7 formulates linear correlation between "potential cyclic fatigue damage" epsilon_p and "equivalent cycling temperature range" DeltaT_e for leadless packages, and equation 9 formulates a quadratic correlation between epsilon_p and DetalT_e for leaded packages. The paper goes on to present more simulated data and experimental data about relative influences on fatigue of solder joints subjected to thermal cycles.

But what are the conclusions in absolute terms for desktop CPUs?

Meanwhile, Intel's "competitor 1" continues to sell us soldered desktop CPUs, of similar die sizes and process nodes compared to Intel's CPUs. This provides us with some empirical life expectancy data of soldered thermal interfaces of typical desktop CPUs. Does it not?

 

[Crumpet]

The result of Hardware Unboxed's 7800X after being overclocked to 4.7ghz @1.25v.

Thats a burnt contact.

https://www.instagram.com/p/BWysDUalr2k/

 

[dullard]

I think those >13yr old soldered Pentium 4s and Ds are running just fine out there. Same for all the 10yr old C2D E6xxx, E8xxx, Q6xxx, Q9xxx, and more recently Nehalems and Sandies.

What was the maximum temperature and typical operating temperature for those CPUs? 65°C to 70°C was quite common back then for maximum temperatures. http://www.pantherproducts.co.uk/posts/cpu-maximum-temperatures/ Actual typical operating CPU temperatures were often in the 45°C range.

Going from 25°C to ~45°C (or even ~65°C) is a far, far cry from the current Intel CPUs going from 25°C to ~95°C. Also, what matters is the speed at which the temperature changes. Intel newest CPUs seem to get hot quite quickly.

 

[dullard]

But what are the conclusions in absolute terms for desktop CPUs?

Meanwhile, Intel's "competitor 1" continues to sell us soldered desktop CPUs, of similar die sizes and process nodes compared to Intel's CPUs. This provides us with some empirical life expectancy data of soldered thermal interfaces of typical desktop CPUs. Does it not?

Until Ryzen is 10 years old, no we do not have empirical data. As for your abolute terms, buy a bunch of CPUs and test it. No reason you have to demand that others do it for you. Solder fatigue is a relatively known science. At least you can ask an expert in that science rather than an online forum (I'm a programmer, not a solder materials scientist).

Until then, we know that small surface areas, large temperature changes, and fast temperature changes are all bad for solder joints. What do Intel chips do? Small surface areas, large temperature changes, and fast temperature changes.

 

[Bouowmx]

Honestly, I thought that the new Intel SpeedShift technology (on Skylake) was supposed to improve the CPU EIST penalty. Not so much.

Can you verify that Speed Shift is enabled, in HWiNFO summary? I have Acer E5-575G with Intel Core i5-7200U, and setting is disabled by default.

See this post about controlling Speed Shift in Windows: http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...s-out-page-501.2428363/page-540#post-38993754

 

[formulav8]

Thats a burnt contact.

Ouch, to much power draw it seems.

 

[Sweepr]

Coffee Lake-S 6C/12T engineer sample @ CPU-Z

Very similar to a previous SiSoftware entry.

Ps: Core i7-8700K is clocked higher.

 

[StefanR5R]

Until Ryzen is 10 years old, no we do not have empirical data.

They did not start to solder with Ryzen. They continued to solder (AFAIK most of) their desktop CPUs during the entire period since Intel started to use TIM in desktop CPUs.

As for your abolute terms, buy a bunch of CPUs and test it. No reason you have to demand that others do it for you.

Nowhere did I demand this from anyone.

Solder fatigue is a relatively known science.

It is possibly even known to Intel's competitor who still solders desktop CPUs.

At least you can ask an expert in that science rather than an online forum (I'm a programmer, not a solder materials scientist).

It was implied in this forum that it is decisive (for desktop CPUs under ambient cooling), so I thought I ask in this forum for further data. I won't do it again.

Until then, we know that small surface areas, large temperature changes, and fast temperature changes are all bad for solder joints. What do Intel chips do? Small surface areas, large temperature changes, and fast temperature changes.

This says nothing about whether or not it is a limiting factor.

 

[scannall]

Lifetime reliability.

http://overclocking.guide/the-truth-about-cpu-soldering/

I've read that more than a few times. And quite frankly, unless you use LN2 it's useless. After over 40 years in this biz, and thousands of CPU's I have never even once seen a CPU failure due to solder on the CPU. Nor even heard of it. As a guess, they lose more dies in packaging using solder than they do using glue. It certainly isn't the material cost itself. At the end of the day, the 'reliability' claim is nonsense. Smoking hot CPU's will die sooner.

 

[dullard]

They did not start to solder with Ryzen. They continued to solder (AFAIK most of) their desktop CPUs during the entire period since Intel started to use TIM in desktop CPUs.

Let me try one more time in as simple of terms as I can make it.

1) Different metals have a different thermal expansion coefficient. The more different this coefficient, the more internal stresses build up with each temperature change.

2) Larger temperature changes (especially quick and large temperature changes) make larger stresses.

3) AMDs chips for years went up to roughly 65°C to 70°C. That limit varied a bit with each chip, but it is roughly true throughout the years.

4) Room temperature is roughly 20°C to 25°C. This will vary obviously from computer to computer due to them being in different environments.

5) Thus, for years you had AMD chips cycling through roughly a 45°C temperature range (25°C to 70°C).

6) Now Ryzen hits 95°C. Thus a Ryzen chip goes through roughly a 70°C to 75°C temperature change.

We have no long-term empirical evidence of AMD chips going through 75°C cycles (almost double the temperature range from what we've seen before). AMD has decided to maximize overclocking capability with solder. Intel has decided not to take the risk of thermal cycles that are nearly twice what we've seen in the past.

Nowhere did I demand this from anyone.

Um, here:

Please elaborate on the die size or/ and design life expectancy where this becomes an issue under ambient cooling.

This says nothing about whether or not it is a limiting factor.

Intel's choice to limit performance when there is a strong competitor says otherwise.

 

[LTC8K6]

The result of Hardware Unboxed's 7800X after being overclocked to 4.7ghz @1.25v.

Thats a burnt contact.

https://www.instagram.com/p/BWysDUalr2k/

Well, we don't really know what caused that. Seems like a lot of people would have this problem, especially reviewers.
That could have been a misaligned pin.

 

[moonbogg]

i7 7800X vs R5 1600. Spoiler: 7800X performs the same and costs twice as much. OUCH!

https://www.youtube.com/watch?v=UfNMn7RWgLw

 

[wildhorse2k]

Well, we don't really know what caused that. Seems like a lot of people would have this problem, especially reviewers.
That could have been a misaligned pin.

Yes most likely bent pin. The burn spot is not centered.

 

[Eddward]

Coffee Lake-S 6C/12T engineer sample @ CPU-Z

Very similar to a previous SiSoftware entry.

Ps: Core i7-8700K is clocked higher.

200Mhz lower base than 8700K and still 4.3Ghz Turbo ? Maybe these 4.5Ghz for 1-2 cores is not impossible on 8700K.

 

[TheF34RChannel]

Coffee Lake-S 6C/12T engineer sample @ CPU-Z

Very similar to a previous SiSoftware entry.

Ps: Core i7-8700K is clocked higher.

Thanks man, just saw this over on Videocardz and what a nice sight it is. My guess is that this is the 8700 (non-K) (which would make sense with the 80W TDP, unless CPU-Z read it wrong, which is also a possibility). I call a 4.6GHz boost on the 8700K (although I expect a 4.5GHz boost actually, but I think you've said it clocked higher(?) so I can see a +100MHz token).

Now @Sweepr; will there be Saturday and Sunday tidbits?

 

[RichUK]

I presume Coffee Lake won't have this mesh rubbish?

 

[TheF34RChannel]

I presume Coffee Lake won't have this mesh rubbish?

No, no mesh mess Good old reliable ringbus and lower than L3 L2 cache.

 

[TheGiant]

i7 7800X vs R5 1600. Spoiler: 7800X performs the same and costs twice as much. OUCH!

https://www.youtube.com/watch?v=UfNMn7RWgLw

well we all know skl-x is not a top game performer- they should try also some of the professional workloads, where IMO 7800X oced to 4.7GHz competes with 8Core ryzen, not 6 Core

They should conclude a 6850K oced to 4.4GHz to see the difference

I don't see the mesh frequency there

And ofc its another review that shows the 7700K is in another gaming league....

But I agree its not a good picture for intel HEDT. Gaming wise, it is a pure fail.

 

[Arachnotronic]

I presume Coffee Lake won't have this mesh rubbish?

Nope, CFL is just a KBL with two extra cores and 4MB extra L3 cache.

 

[swilli89]

well we all know skl-x is not a top game performer...

And ofc its another review that shows the 7700K is in another gaming league....

Actually if Gears of War 4 is removed, 7700k is actually just about 6-7% faster than the Ryzen 1600 chip. Given its 20%+ frequency when overclocked this really isn't that impressive. Depending on what you play as a gamer it might even be the same speed or slower in a lot of titles. 7% overall is not what I would call a "different league".

 

[Arachnotronic]

Actually if Gears of War 4 is removed, 7700k is actually just about 6-7% faster than the Ryzen 1600 chip. Given its 20%+ frequency when overclocked this really isn't that impressive. Depending on what you play as a gamer it might even be the same speed or slower in a lot of titles. 7% overall is not what I would call a "different league".

So, what have you bought to upgrade your Sandy Bridge system that you said a while back that you wanted to replace?

 

[swilli89]

So, what have you bought to upgrade your Sandy Bridge system that you said a while back that you wanted to replace?

Still waiting. For now SNB is enough. And since I sold off my GPU all I really wish I had was a stronger iGPU. I may seriously be looking at building a much smaller system once the better APU of Raven Ridge vs Coffee Lake is decided. Which do you think will perform better in games with only the iGPU?